2025/04/14 by Ignacio Loaiza, Stepan Fomichev, Loaiza, Ignacio +12 · 5 citations
Computer Science · Materials Science · Physics and Astronomy · #Quantum Computing Algorithms and Architecture #Machine Learning in Materials Science #Mechanical and Optical Resonators
paper · pdf · doi:10.48550/arxiv.2504.10602
Near-infrared (NIR) spectroscopy is a non-invasive, low-cost, reagent-less, and rapid technique to measure chemical concentrations in a wide variety of sample types. However, extracting concentration information from the NIR spectrum requires training a statistical model on a large collection of measurements, which can be impractical, expensive, or dangerous. In this work, we propose a method for simulating NIR spectra on a quantum computer, as part of a larger workflow to improve NIR-based chemical detection. The quantum algorithm is highly optimized, exhibiting a cost reduction of many orders of magnitude relative to prior approaches. The main optimizations include the localization of vibrational modes, an efficient real-space-based representation of the Hamiltonian with a quantum arithmetic-based implementation of the time-evolution, optimal Trotter step size determination, and specific targeting of the NIR region. Overall, our algorithm achieves a O(M2) scaling, compared with the O(M12) coming from equivalent high-accuracy classical methods. As a concrete application, we show that simulating the spectrum of azidoacetylene (HC2N3), a highly explosive molecule with strong anharmonicities consisting of M = 12 vibrational modes, requires circuits with a maximum 8.47 x 108 T gates and 173 logical qubits. By enhancing the training datasets of detection models, the full potential of vibrational spectroscopy for chemical detection could be unlocked across a range of applications, including pharmaceuticals, agriculture, environmental monitoring, and medical sensing.